Originally Published in: Forbes
April 9th, 2018
Data is hailed as the marketer’s Holy Grail for a good reason — providing marketers with the insight needed to tailor advertising campaigns helps them maximize engagement among target audiences and return on investment (ROI).
However, harnessing data effectively isn’t always straightforward. For many marketers, the abundance of data produced by disparate sources has made the task of identifying and unifying relevant insight seem colossal. Machine learning, which can take control of data and use it to adjust activity, often in real time, has come to be known as the solution to the industry’s analytical woes.
Machine learning alone, however, cannot offer accurate intelligence for all marketing efforts — data discrepancies have been plaguing the industry for years and we still don’t seem to be any further forward. No matter how sophisticated the technology is, quality insight and results depend on the quality of what is input initially. Research we carried out for a specific advertiser revealed that two critical elements of campaign insight — attribution and performance data — only matched half of the time. The standard of data fed into the machines was therefore only around 50% accurate, meaning the information driving marketing decisions was likely to be equally imprecise. What’s more, this is not a one-off — it’s a widespread anomaly in digital advertising.
The question is: What causes this fragmentation and how can marketers ensure intelligent tools provide a view of the truth that is clear and concise, rather than clouded and contradictory?
The Chief Cause Of Discrepancies
In short, the problem is that data produced by technologies which track advertising impact to inform media buying, such as demand-side platforms (DSPs), are frequently out of sync with data produced by the ad servers which marketers use to measure campaign performance (such as Google’s DoubleClick).
To better grasp this discrepancy, it’s important to understand how data is gathered along the customer journey. Let’s look at an example — before a customer orders from Uber Eats using the app on their smartphone, they see an Uber Eats digital outdoor screen ad while walking through a tube station. Later, while waiting for the bus, that same customer notices an Uber Eats discount ad while scrolling social media on their smartphone. At that point, they return to the app, read some reviews and finally place their order. For marketing purposes, the details of this customer journey are valuable and when the confirmation page loads, it contains conversion pixels that trace all customer touchpoints, from the DSP to the ad server.
The DSP pixel aims to establish how consumers interacted with ads and assign proportional credit for the purchase, but it only analyzes its own media activity and the data scope is therefore limited. By contrast, ad servers have a broader view that covers search activity, display and other media buys. As a result, ad servers are able to generate a more holistic view of campaign performance and are, therefore, typically more reliable. It’s no surprise that many clients use ad server data as their currency.
Keeping Machines Free Of Bad Data
The consequence of this data discord is that marketers who rely heavily on DSP insight to power machine learning algorithms may be feeding them incorrect data. Going back to the Uber Eats customer — the DSP might give all acclaim for the final purchase to the digital outdoor ad on the tube and ignore the influence of other touchpoints entirely. Thus, marketers only using this data to inform future activity may move budgets in the wrong direction, thereby restricting campaign efficiency and ROI.
To maximize advertising results, it’s clear marketers must acknowledge the possibility of incorrect data and put measures in place to improve quality outcomes. Here are two key steps they need to take:
1. Stop depending on DSPs. As we have seen, there is a significant risk that machine learning algorithms fed mostly on DSP data will cap or even derail campaign performance. It is vital for marketers to fuel their tools with comprehensive data that takes all media activity and every touchpoint into account, such as those available via ad servers.
2. Scrutinize each campaign. With no means of predicting how great discrepancies will be, it’s crucial that marketers compare data on a case-by-case basis. Only by closely evaluating the differences across datasets can they precisely pinpoint where discrepancies lie and decide which information is useful and which isn’t. Much of this assessment will need to be carried out by smart tools that can handle large data quantities, but the importance of human insight shouldn’t be underestimated. As well as efficient technology, the marriage of man and machine is crucial for success. Ultimately, marketers must utilize human capacity to tell good data from bad.
Machine learning can transform advertising efficacy, allowing marketers to reconcile data and use these insights to fine-tune campaigns for optimal efficiency, engagement and ROI. The technology isn’t infallible though. While the tech of the fourth industrial revolution is intelligent, modern machines still need high-quality input to work well. So, to avoid hindering campaigns and wasting spend, marketers need to carefully analyze the performance data they use and ensure machine learning is running on the best possible fuel.